# Is this set of vectors representing direction in a sequence trainable across LSTM? if not, what are the drawbacks?

I am starting my Neural Network Project for the first time, I have a lot of expectations:

The main idea is primarily analyzing the pattern of my writings or drawings when using the mouse given a determined set of steps. I recorded myself writing the letter 'Z' and saving the direction of each stroke to a dataset of shape (200, 50, 9). Example of an individual set:

(50, 9)

[[0, 0, 0, 0, 0, 0, 0, 1, 0], ..., [0, 0, 0, 0, 0, 0, 0, 0, 1]],

Where one (1, 9) vector is which direction the mouse is currently heading towards.

My network is fixed just like this:

_________________________________________________________________
Layer (type)                 Output Shape              Param #
=================================================================
lstm (LSTM)                  (None, 10, 50)            12000
_________________________________________________________________
dropout (Dropout)            (None, 10, 50)            0
_________________________________________________________________
lstm_1 (LSTM)                (None, 50)                20200
_________________________________________________________________
dropout_1 (Dropout)          (None, 50)                0
_________________________________________________________________
dense (Dense)                (None, 30)                1530
_________________________________________________________________
dropout_2 (Dropout)          (None, 30)                0
_________________________________________________________________
dense_1 (Dense)              (None, 9)                 279
=================================================================
Total params: 34,009
Trainable params: 34,009


I divided each set into 10 sequences to make the LSTM work. and consecutively the output is the next (1, 9) vector to which the mouse should move.

However, I have hit a wall and the model does not seem to learn, I might suspect there is something wrong with my dataset. So I have come to clarify one thing:

1. Is this dataset even trainable?, or in other words, I want to know whether there is an X model that is capable of predicting the next vector direction. So I just can keep trying and finding the optimal model or even keep cleaning the data. Or, I am missing a basic concept that makes the dataset lack information to be understandable to any model.

Thanks and regards.